The recent competitive successes of a bilateral, transtibial amputee sprint runner who races with modern running prostheses has triggered an international controversy regarding the relative function provided by his artificial limbs. Here, we conducted three tests of functional similarity between this amputee sprinter and competitive male runners with intact limbs: the metabolic cost of running, sprinting endurance, and running mechanics. Metabolic and mechanical data, respectively, were acquired via indirect calorimetry and ground reaction force measurements during constant-speed, level treadmill running. First, we found that the mean gross metabolic cost of transport of our amputee sprint subject (174.9 ml O2·kg−1·km−1; speeds: 2.5–4.1 m/s) was only 3.8% lower than mean values for intact-limb elite distance runners and 6.7% lower than for subelite distance runners but 17% lower than for intact-limb 400-m specialists [210.6 (SD 13.2) ml O2·kg−1·km−1]. Second, the speeds that our amputee sprinter maintained for six all-out, constant-speed trials to failure (speeds: 6.6–10.8 m/s; durations: 2–90 s) were within 2.2 (SD 0.6)% of those predicted for intact-limb sprinters. Third, at sprinting speeds of 8.0, 9.0, and 10.0 m/s, our amputee subject had longer foot-ground contact times [+14.7 (SD 4.2)%], shorter aerial [−26.4 (SD 9.9)%] and swing times [−15.2 (SD 6.9)%], and lower stance-averaged vertical forces [−19.3 (SD 3.1)%] than intact-limb sprinters [top speeds = 10.8 vs. 10.8 (SD 0.6) m/s]. We conclude that running on modern, lower-limb sprinting prostheses appears to be physiologically similar but mechanically different from running with intact limbs.

  • prosthetics
  • running economy
  • sprinting
  • running
  • fatigue
  • locomotion
  • biomechanics

prosthetic legs have existed for millennia, but even today's most advanced models generally do not provide full biological function. The recent athletic performances of a bilateral, transtibial amputee sprinter indicate that the long-standing assumption of functional inferiority may no longer be valid. This amputee athlete has had extraordinary success while racing with prosthetic limbs over the last several years. He narrowly missed the automatic qualifying standard for the 400-meter dash at the able-bodied 2008 Olympic Games. He also finished second in the able-bodied National Championships of South Africa in 2007. These unprecedented achievements for an amputee athlete have raised a provocative question about relative limb function: are modern running prostheses now equal or perhaps superior to biological limbs?

An international scientific and athletic controversy has arisen over this intriguing question. The controversy is rooted at least in part in the limited understanding of the mechanical and physiological consequences of running with prosthetic vs. biological limbs. Here, we present three experimental comparisons between this amputee athlete and competitive runners with intact limbs. Our general objective was to evaluate whether running with lower-limb prostheses vs. running with intact, biological limbs is functionally similar or not. For this purpose, we tested three hypotheses at the whole-body level that would provide relevant, straightforward comparisons: the metabolic cost of running, sprinting endurance, and sprinting mechanics. Conversely, we avoided estimations of whole-body and joint mechanical power and energy transfers because their interpretation is ambiguous (32, 33, 39) and their relationship to sprint running performance is not well understood.

Although there are many informative running studies on unilateral amputee runners (5, 6, 9), the scientific literature contains little information on bilateral amputees (4). The extremely limited, directly applicable information on bilateral, transtibial prosthetic running led us to rely largely on established mechanistic relationships and reasoning to formulate our three hypotheses. First, we assumed that the absence of lower-limb musculature would result in smaller muscle volumes being active during prosthetic running. Accordingly, we hypothesized that the metabolic cost of running with bilateral, transtibial prostheses would be lower than for running with intact limbs. Second, given that mechanical running prostheses do not fatigue, we hypothesized that bilateral, transtibial prostheses would allow a greater proportion of the athlete's top sprinting speed [i.e., anaerobic speed reserve (7)] to be maintained during sprint efforts of longer durations. Third, given that passive, elastic prostheses are designed to provide the spring-like function that human lower limbs do during the stance phase of each stride (12), we hypothesized that the mechanics of sprinting at common speeds would be similar for a bilateral transtibial amputee and runners with intact limbs. Specifically, we hypothesized that the magnitudes of the ground reaction forces in relation to body weight and the respective durations of the contact, aerial, and swing phases of the stride would not differ.


Experimental Design

We conducted our evaluations of functional similarity for prosthetic vs. intact-limb running as follows. First, we used existing data to establish the biological variability present among intact-limb runners on each of the three whole-body measures of interest. Next, we acquired the same data on our amputee sprint subject. We then compared the values measured for our amputee subject to an appropriate group of intact-limb runners. If the values measured during prosthetic running fell within the range of values naturally present for runners with intact limbs, we reached a conclusion of functional similarity; if not, we reached a conclusion of dissimilarity. Quantitatively, we evaluated these comparisons by using a conventional criterion for significance (i.e., P < 0.05). We assumed normal distributions about the intact-limb means and thus set our a priori thresholds for functional dissimilarity at differences of two standard deviations (SD) or greater between amputee and intact-limb values. This statistically conventional but conservative threshold was chosen to minimize the risk of a type I error since we only studied one bilateral, transtibial amputee sprinter.

To test our first hypothesis, regarding the metabolic cost of running, we used the range of biological variability for runners with intact limbs from the most comprehensive study in the literature for competitive male runners at the elite and subelite levels (22). Additionally, we acquired metabolic data on subjects who were competitive 400-m runners with best performances similar to our amputee subject. Our first hypothesis was that the metabolic cost of running for our amputee subject would be greater than two SD below the means reported for each of these three intact-limb comparison groups (i.e., elite runners, subelite runners, and 400-m specialists with similar best performances).

To test our second hypothesis, regarding sprinting endurance, we established intact-limb norms using the sizeable database present in the literature for competitive runners (7, 36). These studies indicate that the all-out speeds of intact-limb runners during any trial lasting from a few seconds to a few minutes can be accurately predicted from two variables: the top sprint speed and the minimum speed eliciting maximal aerobic power. If both of these speeds are known, the speed for any all-out trial from 3 to 300 s is provided by: Math(1) where Spd (t) is the speed maintained for an all-out sprint of duration t, Spdaer, also known as the velocity at V̇o2max (10), is the minimum running speed eliciting maximal aerobic power, Spdts is the maximum or top sprinting speed that can be attained for eight consecutive steps (∼2 s), e is the base of the natural logarithm, and k is an exponential constant for running (0.013 s−1) that describes the decrements in speed that occur with increments in the duration of all-out running.

For our sprinting endurance comparisons, we evaluated whether the measured speeds obtained from all-out sprints of different durations conformed to those predicted by Eq. 1. This relationship has previously been shown to predict all-out sprint performances to within an average of ±3% (7, 36). To evaluate functional similarity for this comparison, we used a criterion of twice the standard error of estimate (SEE). The SEE is the most commonly used statistic for comparing actual vs. predicted values and is the statistical and formulaic analog of the standard deviation. The SEE value utilized here was determined from 84 all-out treadmill trials previously completed by seven competitive runners (7).

Our second hypothesis was that our amputee subject would have appreciably enhanced sprinting endurance because carbon fiber prostheses do not fatigue during sprinting as skeletal muscle does (8, 23, 38). This possibility was suggested by his superior relative performances in longer vs. shorter sprint races and his atypically fast closing velocities while racing. We tested this possibility during constant-speed treadmill trials to eliminate the potentially confounding influence of the start and acceleration portions of overground sprint races. The race velocities of our amputee subject vs. intact limb competitors in the second half of his 400-m races on the track led us to expect all-out speeds ∼10% faster than those of intact-limb controls for any all-out efforts lasting longer than 20 s.

To test our third hypothesis, regarding running mechanics, we compared our amputee subject's sprinting mechanics to the mechanics of a group of track athletes with similar top treadmill sprinting speeds. Here also, we set a functional dissimilarity threshold of greater than two SDs from intact-limb control means at the same running speeds for each of the following variables: foot-ground contact times, aerial times, swing times, stance-average, and peak vertical ground reaction forces. Our third hypothesis was that the running mechanics of our amputee subject would be functionally similar to those of intact-limb runners.

Subject Characteristics

Our amputee subject's average mass while wearing his prostheses (combined prostheses mass: 2.50 kg) was 80.0 kg. His height while standing on his running prostheses (Össur Cheetah, category 5) was 1.86 m; his leg length under the same conditions was 1.01 m. The intact-limb subjects tested to evaluate our first hypothesis were competitive male 400-m specialists with personal best times that were within ±2.0 s of our amputee subject [n = 4, mass = 75.3 (SD 3.8) kg]. The intact-limb runners used to evaluate our second hypothesis were competitive runners whose data led to the formulation of Eq. 1 and the anaerobic speed reserve model (7, 36). The intact-limb subjects used to evaluate our third hypothesis were competitive track athletes with top treadmill sprinting speeds similar to those measured for our amputee sprint subject [n = 4; mass = 72.7 (SD 3.7) kg; leg length = 0.97 (SD 0.04) m]. The leg lengths of intact-limb subjects tested for our third hypothesis matched those of our amputee subject to within 4.0 cm. The body masses of these subjects and our amputee subject conformed to the mean ±2 SD reported (76.2 ± 14.0 kg) of elite male 400-m runners (37). Testing took place in the Locomotion Laboratory of Rice University during February and March of 2008. Subjects provided written informed consent in accordance with the Institutional Review Board of Rice University.

Hypothesis Test I: Metabolic Energy Expenditure During Running

Steady-state rates of oxygen uptake were measured using two methods: a computerized metabolic system (Parvo Medics TrueMax 2400, Sandy, UT) and the Douglas bag method using the specific protocol described by Weyand and Bundle (36). Subjects completed a progressive, discontinuous, horizontal treadmill test that consisted of 5- to 7-min bouts of running interspersed with 3- to 5-min rest periods. The test was initiated at 2.5 m/s and terminated when the subject could not complete the prescribed bout duration of 5–7 min while putting forth an all-out effort. Throughout the test, expired air was directed via a one-way breathing valve and tubing through a pneumotach into a mixing chamber. During the last 2 min of each bout, expired air was also collected in meteorological balloons via the exhaust port of the mixing chamber. Bag volumes were determined using a Parkinson-Cowan dry-gas meter with simultaneous temperature determination. Aliquots were drawn from both the mixing chamber and the balloons for analysis of O2 and CO2 fractions using paramagnetic and infrared analyzers, respectively. All values were corrected to stpd conditions.

Rates of oxygen uptake.

Rates of oxygen uptake (ml O2·kg−1·min−1) were averaged over the last 2 min of each steady-state running trial to obtain the value for each trial speed. Measurements from the computerized and Douglas bag methods agreed to within an average of 1.3% (SD 1.2%). The values reported are those acquired from the computerized system. Both amputee and intact-limb 400-m specialist measures were taken at speeds between 2.5 and 4.5 m/s.

Maximal aerobic power.

The maximal rate of aerobic metabolism, V̇o2max (ml O2·kg−1·min−1), was the highest single minute value measured during the final all-out bout of the treadmill test. Spdaer (m/s) was determined from the measured aerobic maximum and the oxygen uptake-speed regression relationship for each subject. The latter was formulated using steady-state oxygen uptake values from only those trials eliciting <90% of the subject's maximal aerobic power.

Metabolic cost of transport.

The oxygen or metabolic energy cost per unit distance traveled (ml O2·kg−1·km−1) was determined by dividing the rate of oxygen uptake by the speed of the trial. To maintain consistency with literature values, no baseline subtractions of resting oxygen uptake were performed. Thus all oxygen uptake rate and transport cost data are gross rather than net values. Throughout this manuscript, we have reported metabolic energy expenditure in units of oxygen uptake rather than in true units of energy. This practice conforms to physiological convention and facilitates comparisons to the large majority of data previously reported for competitive runners.

Functional similarity for the metabolic cost of running was evaluated using metabolic transport costs rather than rates of oxygen uptake for two reasons. First, our original rates of oxygen uptake were acquired at different specific speeds for our amputee subject vs. intact-limb 400-m runners and over different speed ranges vs. the elite and subelite distance runners (22), as well as many of the other literature values. Direct comparisons of the rates of oxygen uptake acquired at different speeds are not valid. Second, because individual metabolic transport costs vary little across speed (10), the most robust and representative single value for the metabolic cost of running for an individual is provided by the average of the transport costs obtained across a series of steady-state running speeds.

Hypothesis Test II: Sprinting Endurance

In addition to the previously described test to determine the Spdaer, each subject completed a progressive discontinuous treadmill test to determine their top speed. Subjects also completed a total of 6–15 constant-speed, all-out treadmill trials at speeds selected to elicit failure at durations ranging from 3 s to 5 min. During individual test sessions, the number of all-out trials completed ranged from two to five in accordance with previous descriptions (7, 36). Each all-out trial was initiated by the subject lowering himself from the handrails onto the treadmill belt after it had fully accelerated to the desired speed. Subjects were instructed to terminate the run when they were physically unable to match the speed of the tread. They did so by grabbing the handrails and straddling the belt until it was stopped.

Hypothesis Test III: Sprinting Mechanics

Subjects tested to evaluate our third hypothesis completed progressive, discontinuous, horizontal treadmill tests to assess their running mechanics and to determine their eight-step top speed as previously described (35). Tests were completed on a custom, high-speed force treadmill (AMTI, Watertown, MA). The treadmill has a belt width of 0.610 m and is powered by a Baldor motor, and the treadmill bed (0.686 × 2.083 m) serves as a strain gauge-based force platform. Subjects were strapped into a harness secured overhead and slackened sufficiently to become taut only in the event of a fall. Each trial was initiated by the subject lowering himself from the handrails onto the treadmill belt after it had fully accelerated to the desired speed. Our amputee and control subjects were all generally able to transition quickly from standing to running without losing their balance. In the few instances in which these transitions were not made rapidly, subjects were immediately instructed to dismount the treadmill, recover, and prepare for another attempt. These treadmill tests started at speeds of 2.0 to 2.5 m/s. Speed increments ranged from 0.5 to 1.0 m/s through roughly 80% of the subject's estimated top speed, after which speed increments were reduced to 0.1 to 0.4 m/s. The magnitude of each increment was selected in accordance with subjects' performance results on the previous trial and their verbal feedback regarding difficulty. All subjects completed trials at 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, and 10.0 m/s except one of the four intact-limb sprinters. Slower and intermediate speed trials lasted from 10 to 30 s, whereas faster speed trials lasted from 2 to 10 s. Subjects were encouraged to take as much rest as needed for full recovery between trials.

Top speed.

Top speed (m/s) was defined as the fastest speed at which the subject was able to complete eight consecutive steps without backward drift on the treadmill. This was determined by administering trials at progressively faster speeds until a speed was reached at which the subject was unable to match the belt speed for the requisite number of steps while putting forth a maximal effort. Each subject failed on a minimum of two all-out attempts before the test was terminated. In all cases, the top speed successfully completed was within 0.2 m/s or less of the subject's failure speed.

Treadmill force data.

Force data for each trial were acquired using AMTI NetForce software after signal amplification and digitization (DigiAmp, AMTI). Data at each trial speed were acquired at 1,000 Hz and subsequently processed with custom software that applied a Butterworth filter with a low-pass cut-off frequency of 30 Hz (Igor Pro:IFDL, Wavemetrics, Lake Oswego, OR). The values reported for each speed represent means determined from a minimum of eight consecutive steps. Values at three speeds for one of the four intact-limb subjects were interpolated; in each case from measures taken within 0.3 m/s of the interpolated speed. Representative traces from our amputee and one intact-limb subject appear in Fig. 1.

Fig. 1.

A: tracings from video images of our amputee subject during the contact, aerial, and swing phases of a stride while sprinting on a treadmill at 10.5 m/s. Also shown are the vertical (B) and horizontal ground reaction forces, normalized to body weight (C) vs. time for our amputee and an intact-limb subject at a treadmill speed of 10.5 m/s. Black (amputee sprinter) and gray lines (intact-limb sprinter) illustrate the ground reaction force traces of the right (solid) and left limbs (dotted) of the 2 subjects. Differences in the duration of the aerial, swing, and total stride times (shorter) for our amputee vs. intact-limb subject correspond to the dashed line extensions of the respective lines in A (bottom); differences in the duration of the contact time (longer) for our amputee vs. intact-limb subject correspond to the solid line extensions.

Average vertical ground reaction force.

The average vertical ground reaction force applied during the contact or stance phase was determined from the time during which the vertical force signal continuously exceeded a threshold of 40 N. Forces were expressed as multiples of body weight (Wb) by dividing the force recorded during each trial by the weight of the subject recorded on a platform scale before treadmill testing.

Contact time.

The time of foot-ground contact (s) was determined from the periods during which the vertical treadmill reaction force continuously exceeded 40 N.

Aerial time.

Aerial times (s) were determined from the time interval between the end of foot-ground contact with one limb and the beginning of foot-ground contact with the other limb.

Swing time.

Swing time (s), or the time taken to reposition a single limb, was determined from the time elapsing between the end and subsequent beginning of foot-ground contact periods by the same limb. The swing period includes two aerial periods as well as the contact period of the contralateral limb (Fig. 1).

Stride time.

Stride time was determined from the time elapsing between the first instance of contact for consecutive foot strikes by the same limb.

Leg length.

Leg length (m) was measured from the axis of rotation of the right hip joint to the ground at the outside of the right heel or prosthesis blade during erect standing. Hip joint axis of rotation was determined by palpation as the subject slowly swung the limb in the sagittal plane.


Differences in mean values obtained from our amputee sprinter and intact limb subjects are reported as percentages [(amputee sprinter − intact limb)/intact limb]·100 and as multiples of the intact-limb SD or SEE. Unless otherwise stated, data are presented as mean (SD).


Hypothesis Test I: Metabolic Energy Expenditure During Running

Rates of oxygen uptake for our amputee sprint subject increased from steady-state values of 26.5 ml O2·kg−1·min−1 at a treadmill speed of 2.5 m/s to 43.3 ml O2·kg−1·min−1 at the fastest steady-state speed of 4.1 m/s and were well described by a linear fit (V̇o2 = 10.6·speed − 0.45; R2 > 0.99; Fig. 2A). Over the same range of speeds, rates of oxygen uptake for intact-limb 400-m specialists increased from 32.7 (SD 1.5) ml O2·kg−1·min−1 at 2.5 m/s to 50.4 (SD 3.9) ml O2·kg−1·min−1 at 4.1 m/s, a relationship also well described by a linear fit (V̇o2 = 11.1·speed + 4.9; R2 > 0.99).

Fig. 2.

A: mass-specific rates of O2 uptake (V̇o2) vs. treadmill running speed for our amputee sprint subject and runners with intact-limbs. Included are elite distance runners, subelite distance runners, 400-m specialists, and World Cross-country champion Z. Tadese. B: mass-specific V̇o2 expressed per unit distance traveled (running economy) for our amputee sprint subject (v), elite distance runners (iv), subelite distance runners (iii), and 400-meter specialists (ii). The economy of our amputee subject was within 2 standard deviations (SD; dashed lines) of the means of the elite and subelite groups but more than 2 SDs below the mean of the 400-m specialists. For comparison, the economy of an endurance-trained bilateral, transtibial amputee (i) and 2 World Cross-country champions [J. Ngugi (vi) and Z. Tadese (vii)] are also shown. Data sources are as follows: i, Brown et al. (3); iii and iv, Morgan et al. (22); vi, Saltin et al. (29); and vii, Lucia et al. (19). Note: all values reported are from treadmill running at an inclination of 0% except the Lucia et al. value for Z. Tadese, which was collected at a 1% grade.

The gross metabolic cost of transport for our amputee subject averaged 174.9 (SD 2.2) ml O2·kg−1·km−1 (Fig. 2B) and was virtually constant across the five speeds measured. Our amputee sprint subject's gross metabolic cost of transport was 3.8% lower than the mean for elite male distance runners [181.9 (SD 9.1) ml O2·kg−1·km−1 (22)], 6.7% lower than the mean for subelite distance runners [187.5 (SD 9.7) ml O2·kg−1·km−1 (22)], and 17.0% lower than our 400-m specialists [210.6 (SD 13.2) ml O2·kg−1·km−1]. Expressed in terms of the between-subject SDs of the respective groups, the mean transport cost of our amputee sprint subject was, respectively, −0.8, −1.3, and −2.7 × SD lower.

The maximal rate of aerobic metabolism of our amputee subject was 7.6% lower than that of our intact-limb 400-m subjects [52.7 vs. 57.0 (SD 3.4) ml O2·kg−1·min−1; n = 3]. However, he attained essentially the same running speed at maximal O2 uptake (Spdaer) as our intact-limb 400-m specialists [5.0 vs. 4.9 (SD 0.04) m/s] because his metabolic cost of running was relatively lower.

Hypothesis Test II: Sprinting Endurance

The all-out treadmill running speeds in relation to run duration for our amputee sprint subject (Spdts = 10.8 m/s; Spdaer = 5.0 m/s) are shown in Fig. 3A. In absolute terms, these all-out speeds ranged from an eight-step top treadmill speed of 10.8 m/s achieved during a <2.0-s effort to a speed of 6.6 m/s for an 89.5-s effort.

Fig. 3.

A: all-out running speed decreased exponentially in relation to trial duration for our amputee and 3 intact-limb runners of different event specializations. B: relative speed. When the speeds of the 4 runners' all-out trials are expressed as a fraction of their anaerobic speed reserves (top speed − aerobic speed), the fraction maintained at any duration was essentially identical for our amputee and intact-limb subjects. The speeds our amputee sprinter maintained for trials of all durations closely matched those predicted from intact-limb norms (Eq. 1; solid lines in B and C) and fell well within twice the standard error of estimate (dashed lines in C). One of these intact-limb subjects was a sprinter (▾).

For comparative purposes, the data for three intact-limb subjects (one sprinter and two distance runners) also appear in Fig. 3. The all-out running performances of these three intact-limb runners were essentially fully normalized when their trial speeds were expressed as a percentage of their anaerobic speed reserves (Spdts − Spdaer; Fig. 3B). The average agreement between the actual speeds they maintained (n = 35) and those predicted by Eq. 1 was ± 2.1 (SD 2.8)% (Fig. 3C).

When the same anaerobic speed reserve normalization was applied to the all-out performances of our amputee sprint subject, the result was similar (Fig. 3B). The all-out speeds measured matched those predicted from Eq. 1 (using the measured values for Spdts and the minimum speed eliciting maximal aerobic power) to within an average of 2.2 (SD 0.6)% (Fig. 3C).

Thus agreement with the established relationship was essentially the same for our amputee and intact-limb subjects. The all-out speed values for our amputee sprint subject fell within the two SEE (0.50 m/s) prescribed range of functional similarity.

Hypothesis Test III: Sprinting Mechanics

The mechanical means by which our amputee subject increased his running speed from a jog to a fast sprint largely paralleled the patterns observed for intact-limb subjects. The directional changes observed in foot-ground contact times, aerial times, swing times, and stance-averaged vertical force with increasing speed were all similar for our amputee and intact-limb subjects. As treadmill speed was increased from 2.5 m/s to a sprint of 10.0 m/s, foot-ground contact times (Fig. 4A) became progressively shorter. Both aerial (Fig. 4B) and swing times (Fig. 4C) exhibited maximum values at 4.0 m/s and tended to decrease with speed increases from 4.0 to 10.0 m/s. Stance-averaged vertical forces (Fig. 4D) increased sharply from 2.5 to 4.0 m/s but increased relatively slowly from 4.0 to 10.0 m/s. Across the fastest three speeds of 8.0, 9.0, and 10.0 m/s, average vertical ground reaction force increased slightly for intact limb runners but did not increase at all for our amputee sprint subject.

Fig. 4.

Shown are foot-ground contact time (A), aerial time (B), swing time (C), and stance-averaged vertical force (D) vs. speed during constant-speed treadmill running trials for our amputee and intact-limb sprint subjects (n = 4) with similar top sprinting speeds. At the fastest speeds, our amputee subject had longer periods of foot-ground contact, shorter aerial and swing times, and lower stance-averaged vertical forces. Gray shading within the solid lines illustrates intact-limb means ± 2 SDs. All of the mechanical variables illustrated differed between our amputee and intact-limb subjects at the fastest 2 speeds. Vertical force is expressed as multiples of body weight (×Wb).

Although the patterns of change across speed in these four gait variables were similar, magnitudes tended to be less pronounced for our amputee vs. intact limb subjects. Consequently, differences between our amputee and intact limb subjects were minimal at 2.5 and 3.0 m/s, modest at intermediate speeds of 4.0 and 5.0, and appreciable at speeds from 6.0 to 10.0 m/s. At the fastest common speed of 10.0 m/s, our amputee subject's foot-ground contact times were 14.1% longer [0.113 vs. 0.099 (SD 0.004) s], aerial times were 34.3% shorter [0.092 vs. 0.140 (SD 0.011) s], swing times were 21.0% shorter [0.293 vs. 0.371 (SD 0.023) s], and stance-average vertical forces were 22.8% less [1.79 vs. 2.32 (SD 0.10) Wb] than those of intact-limb sprinters. When expressed in intact-limb SD units for each variable, the differences observed at 10 m/s were +3.5, −4.4, −3.4, and −5.2 SD for time of foot-ground contact, aerial time, swing time, and average vertical ground reaction force, respectively. The differences observed at top speed [10.8 vs. 10.8 (SD 0.6) m/s; Table 1] were similar to those observed at 10 m/s.

View this table:
Table 1.

Sprinting mechanics

Horizontal impulses and peak forces were substantially lower for our amputee vs. intact-limb subjects at every speed as shown at 10.5 m/s in Fig. 1. The vertical forces reported throughout the manuscript are therefore conservative in underrepresenting resultant ground reaction force differences between our amputee and intact-limb sprint subjects.


We set out to determine whether near-Olympic-level sprint running performance was occurring via similar or dissimilar physiological and mechanical processes in our amputee and intact-limb subjects. This experimental opportunity was novel but also limited. Sprint running at near-elite speeds with two prosthetic limbs is without precedent and largely unstudied. However, circumstances limited us to testing the one amputee athlete who has these performance capabilities and availed little directly applicable prior information. These limitations might have led to inconclusive results, an inability to distinguish between prosthetic-related and physiological variability, or conceivably both. Yet, the results of all three of our tests were relatively clear. Our first and second hypotheses were primarily physiological comparisons of the metabolic cost of running and sprinting endurance, respectively. Our results indicated that physiological function was largely similar and virtually identical, respectively, between our amputee and intact-limb subjects. The results from tests of our third hypothesis, regarding running mechanics, indicated substantial dissimilarity while sprinting. Accordingly, we conclude that running for our amputee subject is physiologically similar but mechanically dissimilar to running with intact limbs.

A significant concern before testing was the potential difficulty our amputee subject might have performing on the treadmill. A number of factors assured us that this testing apparatus did not hinder his performances in relation to overground running. First, our amputee subject reported being well-habituated to treadmill running from the regular use of his home treadmill. Second, he was able to execute trials of all speeds on our high-speed treadmill in the same manner as our intact-limb subjects did. Third, his sprinting performance during all-out treadmill running at 400-m race speed matched that reported for overground efforts earlier in the off season. Fourth, the metabolic and mechanical data acquired during treadmill running tests on our amputee subject were identical or very similar to those that we obtained during overground running tests. Because virtually all of the intact-limb metabolic and mechanical data available for the three tests undertaken were acquired on the treadmill, we have presented only the treadmill data here.

Hypothesis Test 1: The Metabolic Cost of Running

Because a measurement technique that provides valid estimates for the anaerobic portion and total metabolic energy released during sprinting running has not been developed despite extensive efforts to do so (1, 2, 14, 20, 21, 28), we tested our first hypothesis at the slower speeds required for obtaining valid metabolic data. This was probably not a significant limitation due to the nature of the metabolic rate-running speed relationship. Because this relationship is well described by a linear fit with a near zero-intercept (Fig. 1A), the metabolic cost of transport, or energy expended per unit distance traveled, varies little across speed for different individuals (10, 22).

The results of our first hypothesis test evaluating the metabolic cost of running were mixed. Our amputee subject's costs were lower than the means for intact-limb runners, but only slightly so, being 3.8 and 6.7% (0.8 SD units and 1.3 SD units, respectively) lower than those of elite and subelite distance runners (22). However, his values were 17 (SD 2.7)% lower than those of the intact-limb 400-m specialists tested here and two or more SDs below the means reported for four other groups of subelite male sprinters (24, 25, 31, 34) and 1.67 SDs below those of a fifth group (30). We therefore conclude that our amputee's metabolic cost of running is similar to that of intact-limb elite and subelite distance runners and lower than that of intact-limb male sprinters. However, the differences in the respective metabolic costs incurred by our amputee and intact-limb sprint subjects were largely offset by parallel differences in the aerobic power available to them. As a result, the respective values for the aerobic variable most relevant for sprinting performance, the velocity V̇o2max or Spdaer (7, 36), were nearly identical [5.0 vs. 4.9 (SD 0.04) m/s].

We also note that the metabolic transport cost values that are available for several notable world-class endurance runners with fully intact limbs are lower than those of our amputee subject. These include two world cross-country champions: John Ngugi (29) and Zersenay Tadese (19). Finally, the only other metabolic measurements for a bilateral, transtibial amputee runner (3) that we are aware of, that from a 5-h marathoner, indicated that his metabolic transport costs were 19% greater than our amputee sprinter (Fig. 2B) and similar to nonathletes with intact limbs (22). Without additional data from bilateral, transtibial amputees, a definitive conclusion regarding whether passive-elastic, lower-limb prostheses economize their running is not possible.

Hypothesis Test 2: Sprinting Endurance

The results of our second hypothesis test indicated that our amputee subject's sprinting endurance is virtually identical to that of intact-limb runners. Although his atypically fast closing speeds in races and carbon fiber lower limbs led us to expect a fatigue resistance that would translate into an appreciably greater ability to maintain speed, particularly for those trials lasting as long as 200- and 400-m track events, this was not the case. Rather, we found that our amputee subject's all-out sprinting speeds decreased in relation to trial duration in the same manner as the speeds of intact-limb runners did (Eq. 1). The speeds that we predicted for our amputee subject using intact-limb norms (7, 36) matched those he actually maintained to within 2.2 (SD 0.6)% for six all-out trials between 2 and 90 s in duration.

These results indicate that, when the start and acceleration portions of overground sprint racing are removed, as they were in our constant-speed treadmill trials, the abilities of our amputee and intact-limb sprinters to maintain their sprinting speeds did not differ. Relatively poor starts and accelerations are not surprising for an athlete who lacks ankles, ankle extensor muscles, and feet to transmit muscular force and power distally during the push-off phase (17) of each accelerating step. The slower starts and accelerations of our amputee subject during overground sprint races are likely responsible for his superior performances in longer vs. shorter sprint races relative to athletes with intact legs. Poorer starts and accelerations also inevitably affect pacing by selectively compromising speed in only one portion of a sprint race.

Hypothesis Test 3: Running Mechanics

The results of our third test indicated substantial functional dissimilarity between our amputee and intact-limb subjects in running mechanics. The degree of dissimilarity was almost completely speed dependent: being largely absent at slow speeds, moderate at intermediate speeds, and substantial at the fastest speeds (Figs. 1 and 4). Because running performance at all three Olympic sprint distances is determined primarily by the top sprinting speed of the athlete (7, 36), the mechanics of greatest functional relevance are those that we observed at the fastest speeds.

The speed limits of our amputee and intact-limb subjects were similarly imposed by their gait mechanics. All reached their absolute limit, or top speed, when their foot-ground contact times and vertical impulses decreased to the minimum values necessary to provide sufficient aerial time to reposition the swing leg for the next step (35). Thus, at top speed, our amputee and intact-limb subjects all reached likely maximums for the ground forces they could apply and minimums for the time in which they could reposition their swing legs (Fig. 4, Table 1).

Although the top speed results attained by our amputee and intact-limb subjects were similar, their aerial times, swing times, and weight-specific ground reaction forces were all markedly dissimilar. Given the extent and nature of the mechanical dissimilarities observed, these differences seem largely attributable to running with carbon-fiber, lower-limb prostheses vs. intact limbs. We have previously noted that minimum swing times differ little at the top speed of intact-limb runners of different sprinting abilities, for example, varying by only 0.03 s between runners with top speed of 11.1 vs. 6.2 m/s (35). However, our amputee subject was able to reposition his swing limbs almost 0.10 s more rapidly than the mean we previously reported [0.373 (SD 0.03) s], and 0.075 s, 21%, and 4.0 SD more rapidly than the intact-limb sprinters tested here (Table 1). The combined mass of our amputee subject's residual limb distal to the knee and that of the Cheetah prosthesis is roughly half the mass of an intact calf and foot (4). Reducing the mass of the distal segment of the limb by nearly twofold apparently allows the swing limb to be repositioned appreciably more rapidly.

With his relatively shorter aerial (−34.4%) and swing times and longer contact times (+14.2%), our amputee subject was able to attain the same top sprinting speeds as our intact-limb subjects with stance-averaged vertical forces that were 22%, 0.46 Wb and 3.6 SD units lower than those of intact-limb sprinters. These large force differences at Spdts also seem attributable to sprinting with lower-limb prostheses rather than intact limbs. Transtibial amputees lack the uniarticular, biarticular, and polyarticular muscles that cross one or more of the metatarsal-phalangeal, ankle, and knee joints of an intact limb. The specific absence of bi- and polyarticular muscles disallows the transfer of muscular force possible from the knee to the ankle and foot of an intact limb (17). The lesser ground reaction forces observed in the prosthetic vs. intact-limbs of unilateral, transtibial amputees (11) provide direct evidence of a force impairment.


Perhaps our most striking result, given the interdependence of locomotor physiology and mechanics (18, 26, 27, 32), is that our amputee subject could be simultaneously similar to intact-limb runners physiologically but dissimilar mechanically. Physiological similarity is most likely explained by the reliance of both transtibial amputee and intact-limb runners on the large groups of extensor muscles that act across the hip and knee joints. There was no a priori reason for us to expect that the lower limb prostheses of our amputee subject would alter either the metabolic cost of force production (18, 27) or fatigability (7, 8, 38) at the tissue or fiber level in these skeletal muscles. However, running with lower-limb prostheses might have substantially altered the nature of their activity. Our finding that the whole-body manifestations of these respective skeletal muscle properties (running economy and sprinting endurance) were largely similar suggests that the prostheses, to some extent, approximate the spring-like mechanical function that characterizes intact lower limbs. Although the provision of spring-like behavior from limb segments that lack skeletal muscle is not the norm for human limbs, this phenomenon has biological precedent. Through evolution, the distal limb segments of horses, antelope, and ostriches have lost skeletal muscle and come to rely solely on passive-elastic tendons and ligaments to provide spring-like function.

The mechanical dissimilarities observed highlight the functional trade-offs that are perhaps inevitable for artificial vs. biological limbs. The aerial and swing time reductions observed for our amputee subject support the classic, but largely untested, arguments of functional morphologists. For more than a half century, these scientists have postulated that light, slender limbs have evolved in cursorial animals to enhance speed by reducing the time required to reposition the limbs (13, 15, 16). However, the meager ground reaction forces observed during amputee running here and elsewhere (4, 11) identify what may be a critical limitation for speed (35). Legs must perform different functions during the stance and swing of the stride, as well as during the start, acceleration, and relatively constant-speed phases of sprint running. Collectively, our results underscore the difficulty of providing these multiple mechanical functions with a single, relatively simple prosthetic design and the formidable challenges involved in engineering limbs that fully mimic those produced by nature.


Equipment support and data for intact-limb runners were provided by DAMD17-03-2-0053 from the US Army to P. G. Weyand.


We thank Oscar Pistorius for the exceptional cooperation and openness that made this study possible. We thank all of our subjects putting forth the arduous efforts necessary to complete the full set of experimental tests. We also thank Terry Dial for assistance with the data collection and J. B. Morin for consultation. The law firm Dewey & LeBeouf provided travel and lodging support for Oscar Pistorius and several of the investigators. All of the investigators provided their time pro bono in return for the right to publish these results.


View Abstract